CPC Ensemble-based Extended Range P roducts

1. CPC Ensemble-based Extended Range Products David Unger NOAA/NWS/NCEP Climate Prediction Center

2. Recent Developments • CPC 6-10 day and Week 2 guidance tool based on GEFS reforecasts • Probability of daily extreme temperatures, days 8-14. • Automated “first guess” based on multi-model ensembles.

3. Overview of Ensemble Models Red=Restricted access * = Future product

4. CPC Ensemble Post Processing Model 1 Model 2 Model 3 Direct Model Output Direct Model Output Direct Model Output Bias Correction Bias Correction Bias Correction Probability Estimation Probability Estimation Probability Estimation Consolidation Forecaster Input Public Product

5. Bias Correction Methods • Direct model forecasts (No bias correction) • Bias correction relative to full hindcast data • Adaptive bias correction N-Day trailing running mean Exponentially Weighted Moving Average (EWMA) • PDF Mapping (Hindcast or Adaptive) • Ensemble Regression

6. Bias Correction Method PDF Mapping PDF mapping (PDFc) Model Forecast 20th Percentile Model 20th Percentile Observation Original Model Forecast Bias Corrected Model Forecast Slide Courtesy of Wanqiu Wang

7. Estimation of Probabilities • Direct Member counts (No Bias Correction) • Bias Corrected Member Counts • Kernel-smoothing on regression-corrected forecasts (Ensemble Regression)

8. Regression With Error Estimates Applied

9. Consolidation • Combine ensembles of different models • Combine forecast based on dynamic ensembles with statistical forecast estimates. Methods • Subjective weighting (Blend) • Objective weights Based on the probability of the closest match

10. Best Member Determination Observation Estimated from statistically generated PDF Model 1 PDF Model 1 PDF Accounts for model spread, variable member count Model 2 PDF Accounts for redundancy

11. Products

12. MJO and Teleconnection Indices GEFS, Direct Model Counts

13. Wind Chill and Heat Index GEFS , Bias corrected model counts

14. 500-hpa Heights Consolidation, Direct model forecasts, Manual blend

15. Official Forecast Consolidation, Manual Forecaster adjustments

16. New Products

17. Probabilistic Extremes Product GEFS Ensemble Regression Kernel PDF Selectable percentiles Relative to Climatology Maximum or Minimum Daily Temperature Daily Resolution

18. GEFS Reforecast GEFS, Ensemble Regression, Kernel PDF

19. Bias Corrected Models ECM, Canadian, GFS, Bias Corrected Counts 45-day running means of bias

20. Auto-blend Consolidation, Fixed weights, Subjective blend

21. Experimental Auto Output: Closed-Contoured VGF

22. Influence on Forecast Operations • Well calibrated ensemble based guidance is required for credibility. Badly calibrated reduces forecaster confidence. • Forecasters have (mostly) accepted new guidance. Scores have likely improved. • Consolidation is a major focus. (No clear path forward when models disagree)